Skip to main content
The Tool Calling Editor provides a visual interface for customizing how your agent interacts with tools. Modify input schemas, filter outputs, and configure advanced features like Agentic RAG.

​
Overview

Input Schema

Hard-code parameters, set defaults, or restrict what the agent controls

Output Schema

Filter response fields, remove PII, reduce token usage

Agentic RAG

Semantic querying for large payloads before loading into memory

Custom Instructions

Add context to help the agent use tools effectively

​
Accessing the Editor

  1. Open your agent in the Workbench
  2. Navigate to the Tools section
  3. Click the Settings icon next to any tool

​
Input Schema Editor

The Input Schema tab lets you control the parameters sent to tools.

​
Hard-coding Parameters

Force specific values regardless of AI decisions: 
{
  "repository": {
    "value": "my-org/my-repo",
    "locked": true
  },
  "state": {
    "value": "open",
    "locked": false,
    "default": true
  }
}
  • locked: true - AI cannot override this value
  • default: true - AI can override if needed

​
Use Cases

Hard-code the repository parameter to prevent the agent from accessing other repos:
{
  "owner": { "value": "my-org", "locked": true },
  "repo": { "value": "allowed-repo", "locked": true }
}
Provide sensible defaults the AI can override:
{
  "state": { "value": "open", "default": true },
  "sort": { "value": "updated", "default": true },
  "direction": { "value": "desc", "default": true }
}
Hard-code API keys or tokens:
{
  "api_key": { "value": "sk-...", "locked": true, "hidden": true }
}

​
Output Schema Editor

The Output Schema tab filters API responses before they reach the AI.

​
Selecting Fields

Toggle fields on/off to include/exclude them: 
user:
  - id: true
  - name: true
  - email: false  # Exclude PII
  - phone: false  # Exclude PII
  - department: true
  - role: true

​
Benefits

  • Reduce token usage - Only send relevant data
  • Remove PII - Exclude sensitive fields automatically
  • Focus responses - Help the AI concentrate on important data

​
Agentic RAG Configuration

For tools that return large datasets, enable Agentic RAG to let the agent query semantically.

​
How It Works

  1. Backend API returns large payload (1M+ records)
  2. Agent provides semantic query at runtime
  3. xpander filters using vector search
  4. Only relevant items loaded into context

​
Configuration

agentic_rag:
  enabled: true
  fields:
    - name
    - description
    - content
  max_results: 50
  similarity_threshold: 0.7

​
Example

Without Agentic RAG: 
Tool returns: 50,000 user records
Tokens used: 500,000+
Result: Context overflow
With Agentic RAG: 
Agent query: "engineers in San Francisco"
Tool returns: 23 matching records
Tokens used: 2,300
Result: Precise, relevant data

​
Custom Instructions

Add instructions to help the AI use tools effectively: 
## When to use this tool
Use this tool when the user asks about GitHub issues or pull requests.

## Important notes
- Always filter by state="open" unless user asks for closed issues
- Include labels in the response for categorization
- Rate limit: 5000 requests/hour

## Example queries
- "Show me open bugs" → state=open, labels=bug
- "Recent PRs" → sort=updated, direction=desc

​
SDK Integration

Access tool configurations programmatically: 
from xpander_sdk import Backend

backend = Backend()

# Get tool configuration
tool = backend.get_tool("github-issues-manager")

# Modify input schema
tool.input_schema.set_locked("repository", "my-repo")

# Modify output schema
tool.output_schema.exclude_fields(["email", "phone"])

# Enable Agentic RAG
tool.enable_agentic_rag(fields=["title", "body"], max_results=50)

Tools & Connectors

Overview of tools and MCP connectors

Tools API Reference

Programmatic tool management